The step appliers, frame normalizer and SSE transport move from client/src/hooks/SSE into packages/chat/src/core with their specs, and the client imports them from @librechat/chat. The ask-user-question predicates the content applier reads move with them, and the client's approval module re-exports them so its consumers are unchanged.
* fix: Use Dropdown for Code Execution Settings So Menus Stay Open
* fix: Describe the Workspace Dropdown and Move Select Scenarios to the Code Tool Picker
* fix: Keep a Disallowed Saved Workspace Disabled in the Dropdown
* feat: Scaffold the @librechat/chat package
Adds packages/chat with core, react and components entries, a tsdown build,
a node-environment jest config, turbo and root build wiring, import boundary
rules, design-lint scope, and core types for the chat transport, stream,
host and status. No code moves yet.
* test: Cover the @librechat/chat boundaries as scenarios
* fix: Run the chat suite in test:turbo and build it for the mock lane
test:turbo now selects @librechat/chat, the mock e2e job builds the package the
chat-package scenarios load, and the synthetic repository in the suppression
scenarios carries the packages/chat/src root lint:design:record names.
* chore: Build the chat package in frontend:ci too
* chore: Run the chat suite in test:all
* fix: Hold core TSX to the chat boundary and scan the package for cycles
* fix: Apply the chat import boundary to JavaScript sources
* fix: Keep nested core files off the components entry
* fix: Close the remaining chat package gates and fence aborts by generation
Blocks imports through the @librechat/frontend workspace name, adds the chat
sources to the i18n scan, the CSS color gate and the client Tailwind content,
and lets ChatAbortRequest carry the generation it means to stop.
* fix: Keep JSX out of the chat core and carry abort and reconnect outcomes
Rejects JSX under the core, copies the chat manifest before the Docker
install, keys the client app build caches on the chat sources, fences
reconnects by generation, and types the abort result the transport returns.
* test: Give the CSS gate's scratch tree the chat root it now scans
* fix: Close the remaining @librechat/chat gates and wiring
Rejects import() and require() across the package boundaries, holds chat
components to the literal-copy rule, compiles TSX in the package's jest
transform, builds the package in Frontend Unit Tests, requires its test job for
main promotion, adds it to the unused-dependency scan, and lists it wherever a
sibling workspace is listed: Dockerfile.multi, the docker smoke paths, the
remaining node_modules caches, the reinstall helper, the bun build chain and
AGENTS.md.
* fix: Tighten the chat boundaries and carry steering and Assistants aborts
Template-literal import() and require() and self-imports through the package
name are rejected, user-facing JSX attributes need localized text, the cycle
scan resolves self-imports, Dockerfile.multi builds the package for the client
stage, and the transport carries steer, cancelSteer, armSteer and the
Assistants abort key.
* feat: Match the chat transport and host contracts to the client's
The transport now reports settled and replaced starts, the stream's protocol
version and resumed flag, reconnect replay, generation status, paused-run
resume with approvals or answers, abort and steer protocol versions, and the
request error shape. The host config and auth carry what the client's
ChatSettings now supplies: rename-while-running, model specs, retention mode
and the user.
* fix: Keep lazy React-entry loads out of the core and type the Assistants stop
Relative import() and require() of the /react entry from core are rejected,
and an abortKey stop resolves to the terminal conversation and run messages the
Assistants route returns.
* fix: Fence start controls by protocol and close extension-bearing entry loads
Settled and replaced start results carry the protocol version, core loads of
the /react or /components entry with a file extension are rejected, and
test:packages:chat builds data-provider first.
* feat: Give the Checkbox Theme Roles for Its Fill and Corner
The shared checkbox painted no fill when unchecked and drew the rounded-sm
corner every small control shares, so the ClickHouse theme could not reach
Click UI's checkbox.color.background.default fill or its 0.125rem
checkbox.radii.all corner.
A checkboxRadius appearance role sets the corner and follows radiusSm, the
corner it drew before, when a theme names that and not this. A
checkbox-fill color role paints an unchecked box only when the new
checkboxFillStyle role is fill; the default transparent style keeps every
default theme pixel-identical, and the fill follows surface-primary when a
theme omits it. Checkbox and CheckboxGlyph read both roles.
The ClickHouse theme sets Click UI's fill and corner, and both entries
leave the drift guard's not-expressible list with their probes pinned to
match.
* test: Spell Each Checkbox Scenario Tag Literally
* test: Pin the Theme Cache Digest Built From the New Resolver
* test: Drive useChat Through a Substitute Transport for Send, Stop, Steer, Queue and Errors
* test: Tag the useChat Transport Specs as Acceptance Scenarios
* test: Fail the useChat Specs When a Turn Has No Response Placeholder
* test: Bound the Nested Jest Run and Fail the Scenarios on a Nonzero Exit
* test: Settle a Stopped useChat Turn Only on Its Aborted Final Frame
ChatView was the last chat reader of store.saveDrafts; it now takes the value from useChatSettings(), so a host that supplies its own chat settings drives ask-answer draft saving as it already drives the composer's. Inside the app both sources hold the same value.
* 🧯 feat: Recover Conversations Whose Attachments Keep Failing
Once a branch carries a problem attachment, every turn resends it and
fails the same way. A failed response whose branch carries files now
offers "Retry without files" (re-asks the failed turn with no files as a
new version) and, when earlier turns carry files, "Branch without files"
(a copy up to the last good answer with every upload and code output
left out, with the failed message restored to the composer). File-shaped
errors say so in the card.
The edit view shows a user message's files as removable chips; "Update &
rerun" sends only the kept files and Save is disabled after a removal.
Fork is offered on a failed turn and gains a "Leave out files" option,
backed by a new `excludeFiles` fork option that drops `files`,
`attachments` and the conversation's `files` from the copy.
Bedrock rejects a request whose documents share a name, and history
resends every earlier document, so repeated names are now numbered
before formatting.
* 🧯 fix: Cover Steered Files, Hand Off the Whole Failed Turn, Name Replayed Documents Once
- Excluding files from a fork also strips attachments steered into a
response (`content[].files` on steer parts), which replay like uploads,
and the decision lives in `forkFileScope` in packages/api.
- The recovery card counts steered attachments as files, and reads the
specific code inside a generic `error` envelope, so a rate limit is not
offered files as the fix.
- "Branch without files" hands the failed turn to the copy's composer the
way a message is put back for editing: text through the in-memory
handoff, quotes, manual skills and the reasoning override through their
compose-time atoms.
- Document names are deduplicated over the formatted payload after the
steer stamp, so replayed steer documents are numbered too.
- Fork stays withheld on a failed response, whose copy would carry the
error as an ordinary answer; the card's branch action copies up to the
turn before.
- Remove-file buttons are natively disabled while the editor is locked.
* 🧯 fix: Keep Search Sources, Scope Root Retries, Skip Legacy Rate Limits
- A fork without files keeps metadata-only attachments (search sources,
file search results) and drops only file-backed ones; the recovery card
likewise counts only file-backed attachments as files.
- A turn sent at the root takes none of the conversation's file refs: its
siblings' uploads (an edited first message, or one retried without its
files) no longer come back through the unanchored-continuation fallback.
- Rows that carry a legacy LangChain model, key or rate-limit code are not
offered files as the fix.
* 🧯 fix: Align File Exclusion With Replay's File Shapes, Skip Assistants, Verify the Copy
- Shared `messageCarriesFiles` (with `isFileContentPart`,
`hasContentPartFiles`, `isFileAttachment`) in data-provider names every
shape historical replay reads a file by: uploads, file-backed
attachments, attached `files` on any content part, and `image_file`,
`file` or direct `file_id` locators. The fork drops all of them and the
recovery card detects all of them.
- Assistants conversations get no recovery actions: their upstream thread
keeps every attachment, and those endpoints offer no fork either.
- "Branch without files" checks the returned copy; a server that ignored
`excludeFiles` gets an error toast instead of a false recovery.
* 🧯 fix: Type Replay's Provider-Native File Locators for the Part Predicates
The usage atom families, the breakdown preference and their helpers move from client/src/store/usage.ts to client/src/components/Chat/Input/TokenUsage/store.ts, beside the components that render them, and the store index stops re-exporting them. The SSE usage handler, the token-usage hook, the chat submission hook and their specs import the feature module. No behavior change.
* perf: Mount a sidebar row's menu actions only once its menu is used
ConvoOptions now keeps only its trigger and menu store on a row the pointer has reached; the mutations, route and chat-context subscriptions move into a controller that mounts when the menu opens, the shift quick actions show, or the delete dialog opens, and stays from then on so a request lands in the handlers that made it. A pointer sweeping the un-virtualized Pinned section leaves one trigger per row instead of one full menu.
* test: Watch every menu mutation hook in the mount-cost test
* refactor: Normalize Artifact Type Spellings in One Place
* test: Keep Backticks Out of the Short React Artifact Source
* test: Retry a Timed-Out Sandpack Boot in the Short React Scenario
* fix: Case-Fold Canonical Artifact Types as Well as Their Aliases
* refactor: Define the code highlight cadence default once
The interface schema, the config defaults and the client's normalizer read DEFAULT_CODE_HIGHLIGHT_THROTTLE_MS and MAX_CODE_HIGHLIGHT_THROTTLE_MS from data-provider, and normalizeCodeHighlightThrottleMs moves beside them so its bounds cannot drift from the schema's. A spec pins the schema, config defaults, normalizer and librechat.example.yaml to the same default.
* test: Read the cadence from the signed-in config payload
* fix: Let Field and Focus Call Sites Use the Control Roles
Drop the border-heavy focus rings on the memory value fields, the tool
output details link and the Mermaid dialog controls, so each takes its
primitive's focus-control ring or the global outline. Give
TextareaAutosize a field variant and move the bookmark description onto
it. Drop the quiet border overrides on form controls so they keep the
border-control outline, move defaultTextProps and InputNumber onto it,
and prune the no-restyle counts these call sites no longer carry.
* test: Expect the Edge Focus Treatment on ClickHouse Bookmark Fields
The detail dialog remembers the last control focused inside it and, after each commit, moves focus to the unavailable notice or the dialog title when that control is gone and nothing else took focus. This covers a refresh that drops a conversation starter as well as the existing 403/404 case, without depending on a blur that browsers do not fire consistently for removed elements. Starters are keyed by text and occurrence so the ones that survive a refresh keep their focus.
* fix: Keep Hover-Revealed Row Actions Visible Wherever a Finger Can Reach
* fix: Apply the Coarse-Pointer Reveal to the Composer Palette and Model Menu
* test: Assert the Reveal on an Earlier Reply, Not the Always-Visible Latest One
* test: Read the Timestamp From the Hovered Reply's Own Row
* fix: Hide Until Hover Only Where Hover Exists and No Pointer Is Coarse
* feat: Let a theme reshape the sign-in field and button
The sign-in floating-label field, the submit button and their shared corner read the new authFieldHeight, authButtonHeight and authControlRadius appearance roles, whose defaults reproduce the 44px field, 48px button and 16px corner. The corner follows radius2xl when a theme names only that. The ClickHouse theme draws them at Click UI's single field and button size with the button's square corner.
* fix: Size the sign-in button by its height role alone
A fixed height with vertical padding left a short theme height less room than the label. Specs now expect the role classes on the sign-in field.
* fix: Keep the reset label and password controls inside a themed field height
The reset request label rests in the middle of the field instead of 12px from its top, and the password field's reveal and copy buttons size from the sign-in field height, 4px off its edges and never under 24px. Both keep their stock positions and sizes.
* fix: Hold the password field controls inside the padding it reserves
The reveal and copy buttons stop growing at the stock 2.25rem, so a taller sign-in field keeps them clear of the typed value.
* feat: Let a theme reshape the second-factor and reset-success actions
The 2FA verify and reset-success continue buttons read authActionHeight and authActionRadius, which follow the sign-in button's height and corner when a theme names those and otherwise the Button height and radiusLg they drew, so the stock screens are unchanged and ClickHouse draws them at 32px with its 4px corner.
* test: Reach the second-factor screen through a pending sign-in
* fix: Rest the reset label inside a minimum-height field and keep the action corner on radiusLg
The reset request label rests from an inset that centers it 2px low as before but never past the field's bottom edge, and the other sign-in actions resolve their corner before the sign-in corner is derived from radius2xl, so a theme that names only radius2xl keeps radiusLg on them.
* feat: Adopt the continue icon, artifact timing, avatar border and dropdown intents Tailwind 3 never rendered
Sizes the continue action's glyph to the 19px of the actions beside it, times the artifact sheet's close to the 250ms its unmount waits for and the docked panel's open to 350ms, frames agent avatars with the 1px border they asked for, and lets the dropdown menu primitive cap its height, grow from its trigger and tint a destructive item's icon.
* fix: Cap the dropdown submenu at the space Radix measures
* test: Stop pinning the adopted artifact durations as inert
The sidebar toggle and the Agent Marketplace link use the nav Button variant and the new chat link hovers in surface-nav-hover, so every control in the desktop rail shares the navigation hover the panel buttons and sidebar rows use.
* feat: Route Every Prompt Category Icon Through a Per-Slot Role
* test: Retry Category Icon Reads While the Prompt List Settles
* fix: Carry the Renamed Category Icon Role to Its Slot 4 Name
* fix: Carry the Renamed Category Icon Role Through Every Color Reader
* fix: Keep Renamed Color Roles Out of the Unknown Token Warnings
* refactor: Paint the desktop icon rail with the nav Button variant
The rail's panel buttons now use the shared nav variant, so their hover and selected fills come from surface-nav-hover and surface-nav-selected like the sidebar rows, and the selected state stays on aria-pressed.
* test: Run the rail nav scenarios on a pointer device
* feat: Give Prompt Category Icons a Purpose Color Role
* fix: Keep Token Formatting and Skip Hover on Touch in the Category Icon Scenario
* test: Refresh the Theme Resolver Pin After Rebasing on the Avatar Ink Change
* feat: Open the Pull Request Card When Hovering Anywhere on a Conversation Row
* fix: Address Review Findings on Row Hover for the Pull Request Card
* fix: Cancel Row Hover on Drag and List Scroll
* fix: Limit Row Hover Scroll Cancellation to the Row's Own Scroller
* feat: Filter OpenRouter Reasoning Efforts by the Selected Model
* fix: Address Review Findings on OpenRouter Per-Model Reasoning Efforts
* fix: Report Catalog Failures, Share Lookups and Honor Direct Endpoints for Reasoning Efforts
* fix: Move Capabilities Handler Into packages/api and Harden Catalog Lookups
* fix: Scope Reasoning Capabilities to One Endpoint and Gate Editors Until They Load
* fix: Skip Catalog Reads for Explicit Definitions and Keep Staged Efforts While Loading
* fix: Retry the Reasoning Capabilities Query on Focus and Reconnect After a Failure
* fix: Remember Failed Catalog Lookups, Reject Locked Overrides Early and Keep Restored Efforts
* fix: Hide Efforts Until the Catalog Is Known, Keep Unlisted Models Unknown and Accept Auto
* fix: Mark OpenRouter Endpoints by Their Resolved Host so the Client Fetches Capabilities
* fix: Share One OpenRouter Catalog Rule, Build the Catalog URL Safely and Make the Lifetime Configurable
* fix: Forward Static Catalog Headers, Resolve Relative Next Links and Align Client Freshness
* fix: Limit Capabilities to Exposed Models and Revalidate Against the Server Entry Expiry
* fix: Hide Efforts After a Failed Refresh, Reject Malformed Pagination and Skip User-Scoped Catalogs
* fix: Use Own-Property Lookups, Skip the Catalog for Auto and Revalidate From the Remaining Lifetime
* fix: Treat Request Metadata Headers as Shared, Store Catalogs as Own Entries and Hide Efforts During Refresh
* fix: Keep Configured Query Parameters Across Catalog Pages and Treat Expired Idle Data as Unknown
* fix: Expose Only Configured Variants, Skip Identity Headers on Fixed Lists and Keep Staged Efforts
* test: Pin the Bundled Theme Resolver Output to the Cache Epoch
* test: Pin the Resolver Fallbacks of an Empty Theme Definition
* test: Pin Only the Themes That Can Enter the Cache
* test: Pin an Inline Theme With Role Overrides
* test: Pin the Resolver Output by Role Fixtures and Role Fingerprint
* test: Pin the Persisted Cache Output Per Role and Mode Scope
* test: Generate the Cache Pin Fixtures From the Resolver
* test: Refresh the Cache Pin for the Promoted Stylesheet Roles
* test: Pin Fixture Inputs Apart From Resolver Output, Sort by Code Point
* test: Keep Sample-Independent Output Pinned When Fixture Inputs Move
* test: Pin the Accepted Appearance Values With the Resolver Output
* refactor: Use Spinner Tones, Label Variants and Shared Button Props in Remaining Components
Replaces text colour, round shape and size overrides with the primitives' own tone, variant, shape and size props where the merged classes are identical, drops redundant field overrides, and removes the prose and arbitrary transition classes that generated no CSS or have a token form.
* test: Cover the Quiet Ink of the Marketplace Search Clear Button
* test: Assert the Clear Button Hover Ink Only Where Hover Exists
* refactor: Replace Inline Styles With Classes Outside Chat and the Side Panel
Moves display, overflow, user-select, outline, will-change, backdrop and mask styling onto utility classes and CSS custom properties across Agents, Artifacts, Audio, Bookmarks, Plugins, Projects, Share, routes and the shared client components, and drops the recorded no-inline-styles suppressions they cleared.
* test: Cover Marketplace Grid Columns Driven by the Columns Variable
* test: Pick a Single-File Directory for the Moved-File Prune Fixture
* refactor: Give InputNumber an Option Variant for Setting Values
* fix: Ship the Option InputNumber Reset With the Package Stylesheet
* refactor: Prune the Setting Value Suppressions After Rebase
* refactor: Use Existing Variants and Drop Redundant Overrides in Nav, Skills, Prompts and Endpoints
* test: Cover the Header Menu Trigger in the Default Theme
* chore: Prune Design Suppressions After the Rebase
* feat: Derive the Theme Cache Version and Close the Color Token Drift
* feat: Add a Button Padding Role for the Default Size
* test: Run the Boot Script Through the Version Injection
* fix: Retain Signed-Out Provenance and Tighten Token Declarations
* a11y: Ring Tabs, Table, Dropdown and Menu Items in the Focus Control Role
* test: Tag the Focus Primitives Scenarios Literally
* test: Match the Dark Focus Rule That Yields to Outline Utilities
* refactor: Keep Menu Items and Table Rows on Their Existing Focus Treatment
* refactor: Leave Tabs and Dropdown on the Global Focus Outline
* test: Assert Both Outline Utility Exclusions in the Dark Focus Rule
* refactor: Clear Chat Input Design Suppressions Through Existing Variants
* test: Cover the Chat Composer Send Path After the Suppression Cleanup
* refactor: Keep the FileUpload Action on Its Ghost Pressed Fill